Generate an editor script

Have make_tool draft a safe editor-automation script from a description — you approve before anything is billed.

Repetitive editor work — bulk renames, report generation, consistent asset tweaks — is a good fit for a generated script. This guide walks one make_tool run end to end.

Before you start

Your AI client is connected and supports MCP approval forms (Claude Code does; if yours can't, runs end as cancelled, free of charge).

This is a paid tool — a fixed credit cost per run, charged only when you approve the result. Blocked, declined, and cancelled runs are free. Current price: pricing page.

Describe the task

One or two sentences, with constraints where they matter:

Make me a tool that lists all Blueprint assets unused in the current level,
grouped by folder, with their sizes.

The engine drafts the script and checks it against a strict API allowlist. Unsafe or ambiguous requests are blocked here — you're told why, and charged nothing.

The draft is transformed into a dry-run simulation: destructive operations are commented out and reversible ones are wrapped in editor transactions. Your client shows you exactly what the script would do — this preview is what you're approving.

Approve or decline

Approve — the script is delivered to your chat and the run is billed.
Decline — the run is discarded and the credit hold is released.

Not yet available — live execution of generated scripts is planned as a second approval round. Today the delivered script is always the dry-run transform: inspect it, then run the enabled parts yourself in the editor's Python console.

Run the script

Paste the delivered script into the editor's Python console (Tools → Execute Python Script) or save it under your project's Content/Python/ and run it from there. Re-enable the commented destructive lines only after checking them.

Verify

The delivered script ran in the editor without errors.
The portal's Usage page shows the approved run.

Next steps

make_tool reference — outcomes, costs, and safety model
Security and data

Was this page helpful?